Anovos Product Roadmap
Anovos is built and released as an open source project based on our experience in handling massive data sets to produce predictive features. At Mobilewalla, we process terabytes of mobile engagement signals daily to mine consumer behavior and use features from that data to build distributed machine learning models to solve a wide range of business problems.
On this journey, we faced lots of challenges due to the lack of a comprehensive and scalable library. After realizing the unavailability of such libraries, we designed and implemented Anovos as an open source library for every data scientists’ use.
🛣 The Roadmap
Following the 1.0 release (see the History section below) our main focus for the upcoming incremental releases is on making Anovos even easier to use and more performant. Our goal is to develop Anovos into a tool that can be used on a data scientist's machines as well as state-of-the-art distributed data processing infrastructure.
As we're identifying specific next steps, we'll continuously update this page. If you have any suggestions or feedback, let us know!.
We developed the first fully functional version of Anovos over the course of three major releases: versions 0.1, 0.2, and 1.0.
Version 0.1 (November 2021)
The 0.1 release of Anovos had all the essential data ingestion and comprehensive data analysis functionalities, as well as the data pre-processing and cleaning mechanisms. It also included some key differentiating functionalities, like data drift and stability computations, which are crucial in deciding the need for model refresh/tweak options. Another benefit of Anovos is a dynamic visualization component configured based on data ingestion pipelines. Every data metric computed from the Anovos ingestion process can be visualized and utilized for CXO level decision-making.
- AWS S3 Storage integration
- Read and write to/from local files
- Column selection and renaming
- Support for Parquet and CSV files
- Support for numerical and categorical data types
Data Analyzer and Diagnostics
- Frequency analysis
- Attribute/feature vs. target
- Attribute/feature interaction/association
Data Preprocessing and Cleaning
- Outlier detection (IQR/Standardization)
- Treatment of invalid values
- Missing attributes analysis
Data Health and Monitoring
- Data drift identification (Hellinger Distance, KS, JSD, and PSI)
- Attribute stability analysis
- Overall data quality analysis
Runtime Environment support
- AWS EMR
- Comprehensive 360 degree view report of the ingested data (Numerical & Categorical)
- Executive summary
- Descriptive statistics
- Quality Checker
- Attribute association
- Data drift & stability
Version 0.2 Release (March 2022)
In this release of Anovos, the library supported ingesting data from cloud service providers like Microsoft Azure. The release also added mechanisms to read and write different file formats such as Avro and nested JSON.
The key differentiating functionality of this release is the Feature Explorer & Feature Mapper for data scientists and end-users to resolve their cold-start problems, which will immensely reduce their literature review time.
The V0.2 release also added a capability called Feature Stability estimator based on the composition of a given feature using set of attributes. This will greatly benefit data scientists to understand the potential feature instabilities that could harm the resiliency of an ML model.
With the 0.2 release Anovos was ready to be used in the day-to-day practices of any data scientists or analyst.
- Microsoft Azure Blob Storage integration
- Support for Avro and nested JSON files
- Support for additional data types: Time stamps columns
- Support for Timeseries data ingestion
Data Cleaning and Transformation
- Auto encoders
- Dimension reduction
- Date/Time related transformations
Feature Explorer / Feature recommender (Semantic search enabled)
To recommend potential features based on the industry, use case, and the ingested data dictionary
- Industry specific use cases and respective features
- Supply chain
- Recommendations are enabled by Semantic search capability
- Supported by pre-compiled feature corpus
- This will be an extension of the attribute stability capabilities of the V0.1 release
Extended Spark & Python support
- Compatibility with different Spark & Python versions
- Apache Spark 2.4.x on Java 8 with Python 3.7.x
- Apache Spark 3.1.x on Java 11 with Python 3.9.x
- Apache Spark 3.2.x on Java 11 with Python 3.9.x
Runtime Environment support
- Microsoft Azure Databricks
Version 1.0 (September 2022)
We released version 1.0 of Anovos in September 2022 with all the functionalities needed to support an end-to-end machine learning workflow. Anovos 1.0 is able to store the generated features in an open source feature store, like Feast (see the docs for more information).
It further provided functions to work with geospatial data, running Anovos on the Azure Kubernetes Service, and numerous performance enhancements.